Analysis of factors influencing pancreatic fistula after minimally invasive pancreaticoduodenectomy and establishment of a new prediction model for clinically relevant pancreatic fistula

Data from 432 patients who underwent MIPD between July 2015 and May 2022 in the Third Affiliated Hospital of Soochow University and Fudan University Shanghai Cancer Center were retrospectively acquired and divided into a modeling cohort, an internal validation group and an external verification cohort. The following are the criteria for inclusion in the list: 1. All patients underwent MIPD surgery; 2. No distant metastases were detected by preoperative imaging; and 3. No additional combination resections for malignant tumors were performed. The exclusion criteria were as follows: 1. Incomplete clinical data or loss of follow-up of the patients; 2. Neoadjuvant therapy was administered to patients prior to surgery; 3. Patients with severe underlying diseases who were unable to tolerate surgery; and 4. Patients were solely given palliative care. The Ethics Committee gave their blessing to proceed with this study. The clinical records of all of the patients who participated in this study submitted their signed informed approval, and the data were incorporated into the study. The work has been reported in line with the STROCSS criteria [14].

In this study, a series of perioperative data were collected, including general physiological conditions, preoperative biochemical indicators, intraoperative conditions and histopathological parameters. Age, sex, BMI, smoking history, history of diabetes and the presence or absence of obstructive jaundice, cholangitis, and biliary drainage prior to surgery were among the basic clinical and background information. The preoperative examination included preoperative white blood cells (WBCs), platelets, hemoglobin, neutrophil-to-lymphocyte count ratio (NLR), alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), Na⁺, K⁺, total bilirubin (TB), direct bilirubin (DB), total protein, albumin, triglyceride, cholesterol and international normalized ratio (INR). Histopathology and intraoperative data comprised the American Society of Anesthesiologists (ASA) score, pathologic diagnosis, pancreatic texture, pancreatic duct diameter, operation time, and volume of intraoperative bleeding.

The operations were carried out in accordance with standard protocols and by two surgeons with extensive experience in pancreatic surgery. The feasibility, safety and scope of the operations were determined by imaging examinations before the operation.

For LPD, the surgical procedure was described in our previous article [15, 16]. The patients were placed in the supine position with legs apart. After general anesthesia and tracheal intubation, laparoscopic exploration was conducted to rule out peritoneal and visceral surface metastases. The gastrocolic ligament was divided below the gastroepiploic artery arch with ultrasonic sheers, and the right gastroepiploic artery was transected. The fourth portion of the duodenum and proximal jejunum was mobilized, and the jejunum was transected 15 cm distal to the Treitz ligament with a linear stapler. The stump of the jejunum was dragged into the supramesocolic compartment. The stomach was divided proximal to the antrum with a linear stapler. The superior mesenteric vein (SMV) was identified, and a tunnel was dissected anterior to the SMV and portal vein (PV). The GDA and right gastric artery were clipped and divided. The pancreatic neck parenchyma was divided by an ultrasonic dissector, and the main pancreatic duct was identified and divided with scissors. The uncinate process was exposed at the left posterior aspect of the SMA. The uncinate process was dissected with an ultrasonic dissector along the right border of the SMA, and the tributary vessels were identified, clipped and dissected. The common hepatic artery and the hepatic artery were isolated and hung by tape. Both cholecystectomy and common hepatic bile duct transection were performed, and the corresponding lymph nodes were harvested. The specimens were entirely removed. A duct-to-mucosa pancreaticojejunostomy was performed. An end-to-side choledochojejunostomy was performed using a continuous suture and was strengthened with two stitches at the two corners. The gastrojejunostomy was performed antecolic or retrocolic with a linear stapler, and the common opening was closed to complete the reconstruction. Three abdominal drainage tubes were placed behind the anastomotic site of pancreaticojejunostomy and choledochojejunostomy and at the right side of the inferior vena cava. RPD was conducted with a da Vinci Xi surgical system. The dissection and reconstruction procedures were similar to those of LPD.

A pancreatic fistula is diagnosed when the amylase content of the drainage fluid is greater than three times the upper limit of the normal value of serum amylase at more than three days after surgery, as defined by the International Study Group of Pancreatic Surgery (ISGPS) in 2016 [17, 18]. POPF classification: Biochemical leak (BL) is a past that of a grade A fistula. B-grade pancreatic fistula affects the postoperative process, and the expected postoperative management needs to be changed. Grade C is defined as organ dysfunction affecting one or more organs, reoperation, or postoperative mortality from pancreatic fistula infection. The ISGPS-developed classification rationale and treatment guidelines were used to grade POPF in this study. Grade B and grade C pancreatic fistulas were identified as having clinical significance in this study [17].

For the entire cohort, all data were examined for normality using the Kolmogorov‒Smirnov test. The mean and standard deviation of continuous variables with a normal distribution were calculated, and Student’s t test was used to examine the data. The Mann‒Whitney U test was used to assess non-normally distributed data, which are expressed as the median (interquartile range). The Pearson chi-square test was used to assess the dichotomous variables, and percentages were used to represent the results. In the subsequent multivariate analysis using logistic regression, only the variables with P < 0.05 in the univariate analysis were considered significant. The risk prediction model was created using R software using the factors that had P < 0.05 in the multivariate analysis. The specificity and sensitivity of the model were assessed in this study using the AUC of the ROC and its 95% CI, and the calibration curve was utilized to illustrate the correlation between the observed and predicted incidence. All statistical analyses were carried out by utilizing IBM SPSS Statistics (version 26.0, IBM Corp.) and R Studio (version 4.3.1, Vienna, Austria).

A total of 432 patients who had undergone MIPD were included in this study. Of these patients, 200 were assigned to the modeling cohort, 110 patients were included in the internal validation cohort, and 122 patients were included in the external validation cohort. The cohort for modeling included 112 men, with an average age of 66.1 ± 9.8 years. The internal validation cohort included 68 men, who were on average 66.8 ± 11.0 years old. The clinical data from the two cohorts, including general and biochemical data, did not significantly differ between the cohorts (P > 0.05, Table 1) and were comparable. A total of 16.0% (32/200) of the modeling cohort and 18.2% (20/110) of the validation cohort experienced CR-POPF.

According to the univariate analysis, as indicated in Table 2, BMI ≥ 24 kg/m², WBC > 10 × 10⁹/L, albumin < 35 g/L, triglyceride ≥ 1.7 mmol/L, intraoperative bleeding ≥ 400 mL, pancreatic duct ≤ 3 mm and pathological diagnosis except pancreatic ductal adenocarcinoma (PDAC) or pancreatitis were significantly associated with POPF (P < 0.05). Other factors and the appearance of pancreatic fistula following MIPD were not found to be significantly related.

Multivariate logistic regression was applied in this work to further evaluate the data. At the multivariate level, this study identified six risk predictors of pancreatic fistula after MIPD. The findings indicated that BMI ≥ 24 kg/m² (OR 2.479, 95% CI 1.081–7.762, P = 0.034), albumin < 35 g/L (OR 3.114, 95% CI 1.088–7.283, P = 0.031), triglycerides ≥ 1.7 mmol/L (OR 2.695, 95% CI 1.553–6.034, P = 0.019), intraoperative bleeding ≥ 400 mL (OR 4.695, 95% CI 1.126–9.742, P = 0.025), pancreatic duct ≤ 3 mm (OR 4.145, 95% CI 1.283–7.571, P = 0.008) and pathological diagnosis except PDAC or pancreatitis (OR 3.738, 95% CI 1.167–8.285, P = 0.028) were predictors of POPF (Table 2). Pancreatic fistula and white blood cell count were not significantly correlated with one another (P > 0.05).

The statistically significant risk factors identified by the multivariate analysis, including high BMI, low albumin, high triglyceride, small pancreatic duct diameter, pathological diagnosis except PDAC or pancreatitis, and intraoperative bleeding, were included in the nomogram to help predict the risk of POPF in patients with MIPD. The nomogram showed that a narrow pancreatic duct and substantial intraoperative bleeding had the greatest impact on the score. For each variable, values in the range of 0 to 100 in the chart are proportionally assigned according to the variable’s regression coefficient of its associated with POPF. Through the scale above the model, the individual scores corresponding to the six independent risk factors can be obtained, and one's overall score is calculated by summing all of their individual scores. Patients who have MIPD have a risk of POPF that corresponds to the prediction probability related to their overall score (Fig. 1).

Then, ROC analysis was performed on the prediction model. The modeling cohort's AUC was 0.819 (95% CI 0.747–0.891), and the internal validation cohort's AUC was 0.830 (95% CI 0.747–0.912). The AUC values were all greater than 0.8, indicating that the prediction model had a high-level of discrimination (Fig. 2). In addition, we conducted verifications of the established prediction model. The calibration curve results demonstrated that the mean absolute errors (MAEs) of the two cohorts were 0.018 and 0.03, respectively (the smaller the MAE value was, the higher the calibration degree), demonstrating a greater calibration degree for the prediction model (Fig. 3). On this basis, we also conducted an external validation of the model. The AUC of the external validation cohort was 0.793 (95% CI 0.671–0.915), and the calibration curve showed that MAEs was 0.028 (Fig. 4). The results demonstrated that the actual findings were consistent with the projected risk of POPF.